Chemical Engineering and Processing 43 (2004) 251–262

Design and synthesis of distillation systems using

a driving-force-based approach
Erik Bek-Pedersen, Rafiqul Gani∗
CAPEC, Department of Chemical Engineering, Technical University of Denmark, DK-2800 Lyngby, Denmark

Received 8 November 2002; received in revised form 7 April 2003; accepted 7 April 2003

Abstract

A new integrated framework for synthesis, design and operation of distillation-based separation schemes is presented here. This framework
is based on the driving force approach, which provides a measure of the differences in chemical/physical properties between two co-existing
phases in a separation unit. A set of algorithms has been developed within this framework for design of simple as well as complex distillation
columns, for the sequencing of distillation trains, for the determination of appropriate conditions of operation and for retrofit of distillation
columns. The main feature of all these algorithms is that they provide a simple “visual” method to obtain near-optimal solutions in terms
of energy consumption without rigorous simulation and/or optimisation. Several illustrative examples highlighting the application of the
integrated approach are also presented.
© 2003 Published by Elsevier B.V.

Keywords: Driving force; Distillation; Integrated system; Synthesis; Design; Optimal solution

1. Introduction with reference to the second component forming the bi-

Most separation processes in the chemical industry make
use of some kind of driving force to achieve the desired
separation. It is therefore advantageous to perform a driving
force analysis at the earliest possible stage of the design of
a process. By definition, driving forces exploit differences
in chemical/physical properties between two co-existing
phases in a separation unit. So, if the feed mixture to be sepa-
rated is a homogeneous single-phase solution, the generation
or addition of a second phase is a necessary condition for
the separation to take place. Driving forces can be generated
or caused by various techniques related to different chem-
ical/physical properties such as addition of heat or a force
field. For distillation, the driving force defined in this work as
the difference in composition of a component i between the
vapor phase and the liquid phase is caused by a difference in
the volatilities of component i and all other components in
the system. This driving force is calculated for a binary mix-
ture or a binary pair of key components of a multi-component
mixture. Therefore, the driving force is always calculated
∗ Corresponding author.
E-mail address: rag@kt.dtu.dk (R. Gani).
nary pair. As the driving force approaches zero, separation
of the corresponding key component i from the mixture
becomes difficult, while as the driving force approaches a
maximum, the energy necessary to maintain the two-phase
system is a minimum. This is because the driving force is
inversely proportional to the energy added to the system to
create and maintain the two-phase (vapour–liquid) system.
Therefore, if the distillation design is based on maximizing
the driving force, it should lead to a highly energy efficient
design. Note that this definition of the driving force does
not require the two co-existing phases to be in equilibrium.
The methodology presented here consists of calculation
of the driving force for the pair of key (binary pair) com-
ponents of a binary or multi-component mixture, based on
phase composition data (which may be experimentally mea-
sured or generated through a model). By plotting the driving
force as a function of composition, the location where the
maximum driving force exists can be identified, and based
on this knowledge, design variables such as feed stage lo-
cation, reflux ratio, reboil ratio and number of stages can
be determined, together with estimates of the column com-
position and temperature profiles. Adding the bubble point
pressure curve to the driving force diagram gives the column
pressure [2].

0255-2701/$ – see front matter © 2003 Published by Elsevier B.V.

doi:10.1016/S0255-2701(03)00120-X
252 E. Bek-Pedersen, R. Gani / Chemical Engineering and Processing 43 (2004) 251–262
The objective of this paper is to introduce a framework
of driving-force-based algorithms for the synthesis and de-
sign of separation schemes. This framework comprises al-
gorithms for the design of simple distillation columns (one
feed with two products), complex columns (one or more feed
and/or side products), separation of binary as well as ternary
and/or multi-component mixtures and distillation column se-
quencing (multi-component separations). These algorithms
can be visualized in two dimensions and therefore, they also
provide a useful visualization of the problem solution and are
valid for equilibrium as well as non-equilibrium systems, and
ideal as well as non-ideal and/or reactive systems. The vali-
dation of these visual synthesis and design algorithms is il-
lustrated through comparison of the results obtained through
the use of rigorous models. It is shown that the optimal so-
lutions obtained through the rigorous models not only lie
close to the solution obtained with the driving-force-based
algorithms but also may be used as very good initial esti-
mates for the rigorous-model-based simulations. In this way,
the efficiency and robustness of the rigorous-model-based
solution methods is improved. This paper presents the ap-
plication of some of the new algorithms belonging to the
framework for synthesis and design of distillation columns.
The examples highlight the simplicity, the visual effect as
well as the accuracy of the algorithms.
2. Methodology
The driving force has been defined by Gani and
Bek-Pedersen [3], and is given by
xi βij
Fij = yi − xi = − xi (1)
1 + xi (βij − 1)
As seen in the model equation above, the driving force is
defined as the difference in composition. The terms xi and
yi denote liquid and vapour phase compositions of i, and
Fij is the driving force for component i for property j. The
relative separability βij is a parameter for component i with
respect to property (or separation technique) j, which may
or may not be composition-dependent and provides a mea-
sure of the driving force. Note that in Eq. (1), all the com-
ponent indices are in principle with respect to the second
component k (in the binary pair). However, since the com-
position of the second component k can always be estimated
from that of component i, this index is not used in the Eq.
(1) and the following equations in this paper. The parameter
βij is obtained from a model describing the differences in
composition between two co-existing phases, or measured
composition data. Note in this context that βij = f (T, P,
composition, θ), where θ indicates external factors such as
resistance to mass and heat transfer. From the driving force
model equation (Eq. (1)), it can also be noted that at fixed
P (or T), two-dimensional plots of |Fij | versus xi (or yi ) can
be made where each data point may also indicate a different
T (or P). Therefore, these diagrams can be used to design
and configure separation schemes, including conditions of
operation.
It has already been shown by Gani and Bek-Pedersen
[3] how the driving force diagrams with respect to rela-
tive volatility can be used for near-optimal (with respect
to energy consumption) single distillation column design.
Bek-Pedersen et al. [1] have shown how the driving force
diagrams can be used to obtain near-optimal (with respect
to energy consumption) sequence of distillation trains.
Bek-Pedersen et al. [2] extended and modified these al-
gorithms to allow for the presence of azeotropes in the
multi-component mixtures, to generate hybrid separation
schemes and to allow for scaling in the distillation column
design when “extreme” conditions for the feed mixture
exist, or non-sharp product compositions are desired. Note
that as the location of the feed point on the Dx –Dy line is
determined (see Fig. 1), the corresponding reflux and reboil
ratios are also determined, based upon knowledge of the
products (denoted by A and B). For only one high purity
product, it is necessary to relocate the feed point in order
to achieve the optimal combination of reboil and reflux
ratios. Effects of non-key components are taken into ac-
count through mixture analysis related to mutual solubility,
azeotropic data and the desired separation.
In the proposed framework, six algorithms related to sep-
aration synthesis and retrofit design have been developed.
The first algorithm deals with single-column distillation
and serves as the basis for the development of the driv-
ing force concept, and therefore also the other algorithms.
The second algorithm is for the design of complex distil-
lation columns. The third algorithm deals with pressure
allocation in a sequence of distillation columns. The fourth
algorithm deals with the sequencing of simple distillation
columns for multi-component separations and the corre-
sponding conditions of operation. The fifth algorithm is
for generation of hybrid separation schemes. The last al-
gorithm deals with retrofit design of distillation columns.
The application of these algorithms is dependent on the
condition that two product specifications of the column lie
on either sides of the maximum driving force. Otherwise,
no appreciable difference can be noted when compared
with other methods, as the driving force would not be
maximized.
In this paper all design algorithms (algorithms de-
veloped previously and new ones) have been integrated
in an integrated framework for synthesis and design of
distillation-based separation systems. First each of the al-
gorithms are presented in terms of a step-by-step algorithm.
Algorithm D1 (Single-distillation column design). Algo-
rithm D1 is a base-case design (single distillation column),
which is illustrated through Fig. 1. It is shown here how
the driving force method is applied to obtain the easi-
est separation in a simple visual way, by the determina-
tion of driving force (its size and location) as the design
parameter.
 E. Bek-Pedersen, R. Gani / Chemical Engineering and Processing 43 (2004) 251–262 253

0.3
D
Driving force, FDi
Dy
FDi

D'

Dx
A B
0
0 1
xi
Fig. 1. Driving force diagram for constant βij = 3.

Else, if condition 2c is satisfied, then relocate NF

This algorithm solves the following problem: given a mix-
5% up.
ture to be separated into two products in a distillation col-
Else, if condition 2d is satisfied, then relocate NF
umn with N stages, what is the optimal (with respect to cost
10% up.
of operation) feed plate location and the corresponding re-
8. Apply Eq. (1) (taking the scaling factors determined in
flux ratio for different product purity specifications?
step 4 into consideration) to compute NF for a given value
1. Generate or retrieve from a database the vapor–liquid of N.
data for the binary system in the column. For a
NF = N(1 − Dx ) (2)
multi-component system, select the two key components
to define the “split” (separation task) and use them as
the binary (key) mixture. Algorithm D2 (Design of complex distillation columns (one
2. Calculate the driving force between the two (key) compo- feed and three products)). To solve the design problem for
nents at the actual operating pressure. Plot the calculated one distillation column with three products (side-draw col-
driving force as a function of the light key component umn), the following algorithm is proposed. The algorithm
composition. requires the two driving force diagrams between the two sets
3. Locate the point Dx as the point on the x-axis that cor- of binary key compounds. The algorithm then determines
responds to the largest driving force. the location of the feed stage, as well as the side-draw stage
3.1. In case of an azeotrope, rescale the x-axis and locate and the minimum reflux ratio.
the point Dx as the relative distance between the
two points, one on each side of D, where the driving
force is zero on the x-axis. (The two points are the 1. List the three key compounds according to their boiling
azeotropic point and the bottom/top product.) points.
4. Specify the desired product specifications.
C1 is the lightest boiling,
5. Calculate the minimum reflux ratio (from the slope of
C2 is the intermediate boiling, and
the line BD).
C3 is the heaviest boiling.
6. Determine whether rescaling needs to be applied. If con-
dition 1 or 2 is satisfied, scaling is needed and go to 7.
Table 1
Otherwise, go to 8.
Conditions of distillation column feed and products that require a scaling
7. If condition 1 (see Table 1) is satisfied, go to 7.1. Else factor to be included in the design procedure
condition 2 (see Table 1) is satisfied and go to 7.2.
Condition 1 Condition 2
7.1. If condition 1a is satisfied, then relocate NF between
5 and 10% up in the column. (a) xHK,Z > 0.8 and Dx < 0.7 (a) [(1 − xLK,D )/(1 − xHK,B )] <
Else condition 1b is satisfied, then relocate NF 0.01 and Dx < 0.7
(b) xLK,Z > 0.8 and Dx > 0.3 (b) [(1 − xLK,D )/(1 − xHK,B )] <
between 5 and 10% down in the column. 0.1 and Dx < 0.7
7.2. If condition 2a is satisfied, then relocate NF 10% (c) [(1 − xHK,B )/(1 − xLK,D )] <
down. 0.1 and Dx > 0.3
Else, if condition 2b is satisfied, then relocate NF (d) [(1 − xHK,B )/(1 − xLK,D )] <
5% down. 0.01 and Dx > 0.3
254 E. Bek-Pedersen, R. Gani / Chemical Engineering and Processing 43 (2004) 251–262
2. Generate or retrieve VLE data for components C1 and
C2 , and for components C2 and C3 .
3. Check if C1 and C2 , or C2 and C3 , form an azeotrope.
If yes: stop.
4. Calculate and plot the two driving force diagrams cor-
responding to the two sets of VLE data.
5. Determine which of the two plots exhibits the larger
driving force.
6. Configure the column accordingly.
6.1. If the largest driving force occurs between C1 and
C2 , then the feed should be between the top and
the side draw in the column.
6.2. Else, if the largest driving force occurs between
C2 and C3 , then the feed should be between the
bottom and the side draw in the column.
7. Generate the joint driving force curve such that the
largest total driving force is achieved.
7.1. If the feed is introduced between the top and the
side draw, then the driving force curves should be
joined such that the largest driving force is in the
top of the column.
7.2. Else, if the feed is introduced between the bottom
and the side draw, then the driving force curves
should be joined such that the largest driving force
is in the bottom of the column.
8. Give the number of plates, N, in the column.
9. Calculate the minimum reflux required in the column.
10. Give specifications on the products. Note that the size of
the side draw must be consistent with the overall mass
balance of the column.
11. From the joint driving force curve, determine the
near-optimum position of the side-draw stage.
11.1. Locate the point DS where the two driving force
curves intersect.
11.2. Calculate the near-optimum position of the
side-draw stage from NS = N(1 − DS ), where N
is counted from the top.
12. From the binary driving force plot that exhibits the
largest driving force, locate the near-optimum feed stage
location.
12.1. Locate the point Dx , the position on the compo-
sition axis corresponding to the largest driving
force.
12.2. If the feed is introduced above the side draw,
then calculate the near-optimum feed stage from
NF = NS (1 − Dx ), where NS is counted from the
top. (Note: NS is used instead of N because there
are only NS stages in the sections represented by
the driving force plot.)
12.3. Else, if the feed is introduced below the side draw,
then calculate the near-optimum feed stage from
NF = NS + (N − NS )(1 − Dx ).
Algorithm D3 (Operating conditions). For the determina-
tion of operating pressures in the columns in a distillation
train, the following algorithm has been developed. For appli-
cation of this driving-force-based method for determination
of distillation column pressures, it is necessary that the dis-
tillation column sequence has already been determined. This
could for example have been done by Algorithm S1 in this
paper. Besides, an appropriate thermodynamic model for the
prediction of mixture bubble points must be available.
1. Calculate data for the P–x–y diagram for the two key
components in the first distillation column at the bubble
point temperature of the feed. Set k = 1.
2. Draw the driving force curve from the data calculated in
step 1 together with the bubble point curve.
2.1. Identify the point Dx as the composition xi , where
the driving force reaches its maximum value (Dy ).
2.2. Identify the bubble point pressure at the point Dx .
Allocate this pressure as the operating pressure for
the condenser in distillation column k.
2.3. Based on a specified pressure drop per plate, deter-
mine the reboiler pressure for column k.
3. Calculate data for the P–x–y diagrams for the two key
components in the next distillation column (k + 1) as
a function of the temperature. Identify the temperature
at which the P–x–y diagram gives the maximum bubble
point pressure to within 5% of the reboiler pressure of
column k. Select this temperature as the feed temperature.
4. Repeat step 3 until all the condenser and reboiler pres-
sures have been allocated in the distillation column
sequence.
Note. If the pressure in one or more of the columns in the
distillation train is found to be lower than atmospheric, a
higher pressure in the first column should be considered as
an alternative.
Algorithm S1 (Sequencing of distillation columns). This
algorithm provides a driving-force-based solution to the
problem of synthesizing a distillation train. Input data to
this algorithm is the identity of the compounds in the feed
mixture and vapour–liquid phase compositions. The algo-
rithm determines the NC-1 separation tasks (that is adjacent
pairs to be separated) and the sequence in which they should
occur.
1. List all the compounds in the mixture, NC, according to
their relative separability, βij (or relative volatility).
2. Rank the compounds by normal boiling points. Retrieve
the vapour–liquid data available for each binary pair of
adjacent key compounds in the NC-1 splits.
3. Calculate the driving force diagrams for the binary pairs
of key compounds, all at the same pressure (usually at-
mospheric pressure).
4. In total, NC-1 driving force curves are calculated. Set
k = 1.
If a binary pair forms an azeotrope, multiply the max-
imum driving force for this pair by a penalty factor
 E. Bek-Pedersen, R. Gani / Chemical Engineering and Processing 43 (2004) 251–262
((Dy,min /Dy,max )Dy,azeotrope ) in order to make the value
the smallest among all the binary pairs.
5. For the split k, select the adjacent pair having the largest
driving force.
6. Add a distillation column for the separation task k.
7. Remove the split between the selected adjacent pair from
the list. Set k = k + 1, and repeat the algorithm from step
4 until only one split remains to be allocated. Otherwise,
go to step 7.
8. For each column of the distillation flowsheet, the feed
stage location may now be determined, if desired, through
Algorithm 1.
Note that this sequencing algorithm provides the largest
total driving force for the generated flowsheet. In this way
the overall separation becomes the easiest possible, and will
thus require the least amount of energy input.
Algorithm S2 (Generation of hybrid separation se-
quence). This driving-force-based algorithm solves problems
related to the identification of the optimal separation se-
quence(s) for a given separation task. The phase com-
position data for the compounds to be separated, for
all the separation methods to be considered, are used
as input data (available knowledge). Based on this
data, the algorithm determines the set of feasible solu-
tions and identifies the solution with the largest driving
force.
1. For any binary pair of components that forms an
azeotrope (or eutectic points or exhibits mutual solubil-
ity), retrieve the sets of composition data corresponding
to two co-existing phases. Note that each set of data
corresponds to a different separation technique. Also,
two sets of data at different operating conditions for the
same separation technique are considered here as two
different separation techniques.
0.6
0.5
0.4
Xi Max
0.3
0.2
0.1
0.0
0 2 4 6
Fig. 2. Plot of largest driving force, and the corresponding location of it as function of alfa.
255
2. Calculate and plot the corresponding driving force dia-
grams for all separation techniques (each corresponding
to a property j).
3. For the specified product purities, identify all feasible
paths, allowing a switch from one separation technique to
another if necessary, by moving along the driving force
curves (see Fig. 2). Note: if one separation technique
is unable to achieve the desired separation, switch to
another that is feasible.
4. For each feasible path, identify the corresponding sepa-
ration techniques and operating conditions from the driv-
ing force diagram (using Algorithm S1).
5. Select as the initial flowsheet one with the largest total
driving force.
6. Use the information from steps 4 and 5 to formulate and
solve a structural optimisation problem to determine the
optimal flowsheet (not covered in this paper).
This algorithm formulates an optimisation problem in
the last step and only the solution of the last step may
provide the optimal solution. However, if the theory of
driving force being inversely proportional to energy con-
sumption and separability is correct, the result from step
5 should provide a very good estimate to the optimal so-
lution (from step 6). In step 6, a rigorous model may be
used. This algorithm therefore serves as a tool for formu-
lation of the synthesis/design problem as an optimisation
problem and for providing good initial estimates to the
solution.
Algorithm R1 (Retrofit design of distillation columns). This
algorithm solves retrofit problems where the design of any
existing distillation column is known and it is necessary to
determine if it can be used to separate a specific mixture
(and the corresponding condition of operation) and/or the
set of mixtures together with the conditions of operation
where the use of the distillation column would be feasible.
0.6
Xi Max
Max FDi 0.5
0.4
Max FDi
0.3
0.2
0.1
0.0
8 10 12
Constant Alfa
256 E. Bek-Pedersen, R. Gani / Chemical Engineering and Processing 43 (2004) 251–262

The starting point for this algorithm is the design of an ex- 1. Specify the desired product specifications (purities).
isting column in terms of number of stages and a range of 2. Determine the minimum reflux ratio RRmin (using
locations for the feed. Now, given the desired product spec- Table 2) and actual reflux ratio RR (from Table 2 by
ifications, the objective is to determine which mixtures can selecting a value of C for the known Nideal .
be separated into desired products and the corresponding 3. Use Table 2 to check if for the calculated values of RRmin ,
conditions of operation. On the other hand, if the mixture the specified product purities can be matched. If this is
identity is also known, then the algorithm simply checks if not matched, return to step 2 with another product spec-
the desired separation is feasible. In this way, the retrofit ification.
design of distillation columns is the “reverse” of a distil- 4. Determine the corresponding values of FDimax of the col-
lation column design problem where given the mixture to umn from Table 2 using the known values of number of
be separated, the design of the column and the condition of stages (last column). Use linear interpolation, if neces-
operation is determined. sary. This may give more than one value of FDimax since
a single value of Nideal in Table 2 corresponds to multiple
values of FDimax . This gives an upper and lower bound
Table 2 of FDimax .
Corresponding values of reflux ratio, minimum reflux ratio, number of 5. Determine the range of relative volatilities, a, (from
stages, product purities and driving force
Fig. 2) that corresponds to the calculated values of
FDimax XLK,Dist XLK,Bot RRmin RRmin × C Nideal FDimax . This also produces an upper and lower bound
0.045 0.995 0.005 9.89 14.83 96 for a.
0.98 0.02 9.56 14.36 71 6. For a known mixture to be separated, check if the mix-
0.95 0.05 8.9 13.35 54 ture exhibits this range of a. If no: stop, as the use of
0.90 0.10 8.22 12.33 41 the given distillation column is infeasible for the desired
0.065 0.995 0.005 7.33 11.0 67 separation. If yes, go to step 8. If the mixture to be sep-
0.98 0.02 7.10 10.65 50 arated is not known, go to step 7.
0.95 0.05 6.64 9.96 38 7. Search a database or other knowledge bases in order
0.90 0.10 5.72 8.58 29
to identify a binary pair or a ternary mixture (where
0.101 0.995 0.005 4.50 6.74 44 the third component is a solvent, and calculating the
0.98 0.02 4.35 6.52 33
a on a solvent-free basis) that matches the calculated
0.95 0.05 4.05 6.08 25
0.90 0.10 3.56 5.33 19 a from step 4. Note that in this step, binary mixtures
that match the calculated a and/or solvents that, when
0.146 0.995 0.005 2.94 4.41 31
0.98 0.02 2.84 4.26 23
added to a binary mixture, match the calculated a can be
0.95 0.05 2.63 3.95 18 determined.
0.90 0.10 2.29 3.44 14 8. Apply Algorithm D1 to verify the feed location.
0.172 0.995 0.005 2.35 3.53 27
0.98 0.02 2.26 3.40 20 Note that Fig. 2 may be used together with Table 2 for
0.95 0.05 2.09 3.13 15 retrofit design, when the number of stages and desired prod-
0.90 0.10 1.80 2.70 12 uct purities are known and/or for simulation purposes. For
0.195 0.995 0.005 2.06 3.09 24 example, given the mixture to be separated and the number
0.98 0.02 1.98 2.97 18 of stages, determine the product compositions and the corre-
0.95 0.05 1.82 2.74 14 sponding composition profiles. By following Algorithm R1,
0.90 0.10 1.57 2.35 11
one in principle moves from the right side of Table 1 to-
0.225 0.995 0.005 1.73 2.60 21 wards the left side. One knows the number of stages and
0.98 0.02 1.67 2.50 16
the desired product purities, this then gives the minimum
0.95 0.05 1.53 2.30 12
0.90 0.10 1.37 1.97 9 reflux ratio, which then again gives the driving force avail-
able. Once the driving force is known, the range of relative
0.268 0.995 0.005 1.37 2.06 18
0.98 0.02 1.31 1.97 13
volatility, α, can be found from Fig. 2. In Table 1, a value
0.95 0.05 1.20 1.80 10 for C = 1.5 has been applied, where C is the multiplication
0.90 0.10 1.02 1.52 8 factor of RR to RRmin .
0.382 0.995 0.005 0.82 1.23 13 The algorithms presented above have been integrated into
0.98 0.02 0.78 1.17 10 a systematic framework for synthesis, design and retrofit.
0.95 0.05 0.70 1.05 8 This framework is illustrated in Fig. 3. Here the algorithms
0.90 0.10 0.57 0.86 6 are classified in terms of classes and, within each categories
0.478 0.995 0.005 0.54 0.81 10 (level 1), calculations (level 2). The categories of level 1
0.98 0.02 0.51 0.76 8 define the problem type in terms of synthesis, design and
0.95 0.05 0.44 0.67 6
retrofit. The calculation or level 2 defines the detailed cal-
0.90 0.10 0.34 0.51 5
culations and sub-problems under any category.
 E. Bek-Pedersen, R. Gani / Chemical Engineering and Processing 43 (2004) 251–262 257

Synthesis of
Design of Distillation Retrofit of
Level 1

Distillation Based
Columns Distillation Columns
Separation Schemes

Sequencing of Hybrid Simple Complex Retrofit Design

Operating
Distillation Separation Distillation Distillation of Distillation
Conditions
Columns Schemes Column Design Column Design Columns

algorithm S1 algorithm S2 algorithm D1 algorithm D2 algorithm D4 algorithm R1

Level 2

Given: Given: Given: Given: Given: Given:

Composition xDist, xBot, Compounds in
and Key Experimental P, NP, xDist, xBot P, NP, x Dist, xBot splits, Inlet NP, RR min
Compounds Phase Eq. Data Pressure

Find: Find: Find: Find: Find: Find:

Order of splits in Combination of Operating
the distillation Units to achieve RRmin, NF RRmin, NF, NS Pressure in NF, xDist, xBot
scheme separation Distillation Seq.

Fig. 3. Framework for synthesis, design and retrofit, connecting the algorithms presented.

Level 1: In the first level the character of the problem is de-
termined, and it is assigned to one of the three categories: (1)
synthesis, (2) design and (3) retrofit. While the retrofit cate-
gory only has one sub-problem associated to it, the synthesis
and design categories have more sub-problems associated.
Level 2: In the second level, the solution paths for indi-
vidual synthesis, design and retrofit problems are outlined.
Each problem has a driving-force-based algorithm associ-
ated to it. For the individual algorithm, it is then given
what is required as input to the algorithm, and what is the

Table 3
Data sheet for Example 1
Action

Algorithm S2

Steps 1 and 2 Identify azeotrope and retrieve data for different

separation techniques

Step 3 Each component must be near unity in

composition (thus the azeotrope must be broken).
Identify feasible paths
Step 4 Identify separation techniques for each feasible path Alternatives: (1) Distillation and pervaporation; or
(2) distillation with solvent; or (3) distillation
with solvent and pervaporation
Step 5 Choose combination with largest driving force A hybrid flowsheet of distillation combined with
pervaporation is chosen as initial flowsheet
Validation
Step 1 Identification of alternatives Distillation + pervaporation. Add solvent to distillation
Step 2 Rigorous calculation of energy duties Distillation and pervaporation consumes a
maximum of 73% of the energy for a distillation
column and a solvent recovery column
258 E. Bek-Pedersen, R. Gani / Chemical Engineering and Processing 43 (2004) 251–262

Table 4
Data sheet for example 2a
Action Problem 1 feed (0.05, 0.90, 0.05)

Algorithm 2
0.20 Pentane-
Hexane
0.16
Hexane-
Heptane
0.12
Steps 1–5 Calculate 2 FD curves;
FDi

Dx = 0.45; 0.08
Pentane–hexane exhibits
the larger driving force 0.04

0.00
0.0 0.2 0.4 0.6 0.8 1.0
Composition (x light key)

0.20
Hexane-
0.16 Heptane
III D
Pentane-
F 0.12 Hexane
FDi

II
Steps 6, 7 Generate joint curve. Ds = 0.32 0.08

S
I 0.04

0.00
0.0 0.2 0.4 0.6 0.8 1.0
B
Relative composition

Step 8 Give number of stages N = 36 (given already)

Step 9 Determine RRmin 2.02
Step 10a Give Spec 1 XD (pentane) = 0.998
Step 10b Give Spec 2 XB (heptane) = 0.85
Step 10c Determine side draw Side draw = feed × 0.90
Step 11 Determine NS NS = 36(1–0.32) = 24.48 ∼ 24
Step 12 Determine NF NF = 24(1–0.45) = 13.2 ∼ 13
Validation
23
Reboiler Duty (GJ/h)

21

19

Step 1 Simulation for various NS . NSopt = 23 17

15
17 19 21 23 25 27

Side Draw Stage, NS

17.0
Reboiler Duty (GJ/h)

16.5

16.0

Step 2 Simulation for various NF . NFopt = 14 15.5

15.0
10 12 14 16 18 20

Feed Stage Location, NF

Step 3 Actual RR/V/F 116.7/5.78

 E. Bek-Pedersen, R. Gani / Chemical Engineering and Processing 43 (2004) 251–262 259

Table 5
Data sheet for example 2b
Action Problem 1 feed (0.34, 0.33, 0.33)

Algorithm 2
0.20 Pentane-
Hexane
0.16
Hexane
Heptane
0.12
FDi

Steps 1–5 Calculate 2 FD curves; Dx = 0.45 0.08

0.04

0.00
0.0 0.2 0.4 0.6 0.8 1.0
Composition (x light key)

0.20
Hexane-
0.16 Heptane
III D
Pentane-
F 0.12 Hexane
FDi

II
Steps 6, 7 Generate joint curve; Ds = 0.32 0.08
S
I 0.04

0.00

B 0.0 0.2 0.4 0.6 0.8 1.0

Relative composition
Step 8 Give number of stages N = 36 (given already)
Step 9 Determine Rrmin 2.02
Step 10a Give Spec 1 XD (pentane) = 0.998
Step 10b Give Spec 2 XB (heptane) = 0.85
Step 10c Determine side draw Side draw = feed × 0.33
Step 11 Determine NS NS = 36(1–0.32) = 24.48 ∼ 24
Step 12 Determine NF NF = 24(1–0.45) = 13.2 ∼ 13
Validation
8
Reboiler Duty (GJ/h)

Step 1 Simulation for various NS ; NSopt = 24 6

5
19 21 23 25 27 29

Side Draw Stage, NS

6.5
Reboiler Duty (GJ/h)

6.0

Step 2 Simulation for various NF ; NFopt = 12 5.5

5.0
8 10 12 14 16 18 20

Feed Stage Location, NF

Step 3 Actual RR/V/F 4.91/1.67

260 E. Bek-Pedersen, R. Gani / Chemical Engineering and Processing 43 (2004) 251–262

Table 6
Data sheet for Example 3
Action Problem 1 feed (0.33, 0.33, 0.34)

Algorithm 2
0.16 Benzene -
Toluene

0.12 Toluene - Xylene

FDi

0.08

Steps 1–5 Calculate 2 FD curves; Dx = 0.46

0.04

0.00
0.0 0.2 0.4 0.6 0.8 1.0

Composition, Light Key Compound

0.16

0.12
III D
FDi

F 0.08

Steps 6, 7 II Generate joint curve; DS = 0.38 Benzene -

0.04 Toluene
S Toluene - Xylene
I
0.00
0.0 0.2 0.4 0.6 0.8 1.0
B
Relative composition

Step 8 Give number of stages N = 40 (given already)

Step 9 Determine RRmin 3.28
Step 10a Give Spec 1 XD (benzene) = 0.995
Step 10b Give Spec 2 XB (xylene) = 0.92
Step 10c Determine side draw Side draw = feed × 0.33
Step 11 Determine NS NS = 40(1–0.38) = 24
Step 12 Determine NF NF = 24(1–0.46) = 12.96 = 12/13
Validation
28
Reboiler Duty (GJ/h)

26

24

22
Step 1 Simulation for various NS ; NSopt = 23
20

18
18 20 22 24 26 28

Side Draw Stage, NS

22.0
Reboiler Duty (GJ/h)

21.5

21.0

Step 2 Simulation for various NF ; NFopt = 12

20.5

20.0
8 10 12 14 16

Feed Plate Location, NF

Step 3 Actual RR/V/F 21.7/7.13

 E. Bek-Pedersen, R. Gani / Chemical Engineering and Processing 43 (2004) 251–262
output from the algorithm. In this way it is easy to get a clear
overview of what is needed for applying the algorithms in a
systematic way.
3. Applications
The applicability and scope of the driving-force-based al-
gorithms are illustrated by a number of application exam-
ples. Algorithm D1 has been verified for a great number of
binary and multi-component mixtures, some of which have
been highlighted in [3]. These examples show clearly that
the minimum energy consumption corresponds to the feed
stage located on the relative position of the largest driving
force, as predicted according to Algorithm D1. In all these
examples the energy consumption has been calculated for
the following specified variables, NP , NF , product specifica-
tions (points A, B in Fig. 1) and feed mixture details. Ap-
plications of Algorithms S1, S2 and D3 have been reported
elsewhere (see, e.g. [1] for S1 application examples, and [2]
for S2 and D3 application examples).
In this paper, therefore, examples highlighting the ap-
plications of the most recent developments in the frame-
work for synthesis and design are presented. One example
of Algorithm S2 is also given; otherwise, the focus is on the
algorithms of complex distillation design (Algorithm D2)
and retrofit design (Algorithm R1). In each example an ap-
propriate property model has been used to determine the
driving force as a function of composition, and the algo-
rithms have been followed according to the steps outlined
above. For Algorithm D2, a large number of examples have
been worked out. This is partly due to the fact that the
same systems have been applied for various feed compo-
sitions (see also Example 2). The results obtained from
the Algorithms D2, S2 and R1 have been verified through
rigorous-model-based simulations (using the steady state
Table 7
Data sheet for Example 4
Action
Algorithm R1
Step 1 Determine desired products
Steps 2, 3 Find RRmin
Step 4 Find FDimax
Step 5 Find range of α
Step 7 Check for mixtures with these properties
Step 8 Check for feed location
Verification
Mixture 1 Butane–i-butane, 5 atm
Mixture 2 Cycloheptanol–cyclooctanol, 5 atm
Mixture 3 1,4-Butanediol–1,3-butanediol, 15 atm
Mixture 4 Hexanol–hexanal, 12 atm
Mixture 5 Diethylene glycol–1,6-hexanediol, 3 atm
261
distillation model in PRO/II [4]). The results from the ex-
amples are given in the form of individual data sheets,
where the individual steps of each algorithm can be followed
easily.
Example 1 (Separation of ethanol and water). This is a
well-known example of separation of ethanol and water,
which forms a binary azeotrope. This example serves to il-
lustrate Algorithm S2. In the figure given in Table 3, the
driving force curves for three different two-phase separation
techniques are given. These are then compared throughout
the range of composition for the desired separation, and the
combination of separation techniques exhibiting the largest
total driving force is identified. As illustrated in Table 3, the
combination of distillation with a membrane unit for per-
vaporation clearly is more efficient considering only energy
consumption during operation.
Example 2 (Separation of three hydrocarbons). In this ex-
ample, the separation of pentane, hexane and heptane in
one column is considered. Tedder and Rudd [5] studied
this system in the context of the design of complex dis-
tillation columns. Two scenarios are treated in this exam-
ple: one where the feed mixture contains hexane (90%) and
5% of each of the two other compounds, and one where
the feed mixture is nearly equimolar. In both cases it is
given that the columns operate at 5 atm pressure and have
36 stages. The two driving force diagrams for this system
are given as figures in Tables 4 and 5. It is clearly seen that
the pentane–hexane split is the easier split (in terms of size
of driving force). Therefore this split is the primary split,
which is then followed by a stripping section in the col-
umn to perform the split between hexane and heptane, as
shown in the sketch of the column under steps 6 and 7 in
Tables 4 and 5.
XB,HK = 0.995, XD,LK = 0.995
Rrmin ∼ 6.4
∼0.07
α ∼ 0.33–0.34
(1) Butane–i-butane, 5 atm
(2) Cycloheptanol–cycleoctanol, 5 atm
(3) 1,4-Butanediol–1,3-butanediol, 15 atm
(4) Hexanol–hexanal, 12 atm
(5) Diethylene glycol–1,6-hexanediol, 3 atm
For all the above binary mixtures, the feed location is matched.
α ∼ 1.33–1.34, FDi ∼ 0.074
α ∼ 1.34–1.38, FDi ∼ 0.080
α ∼ 1.3–1.85, FDi ∼ 0.065−−1.45
α ∼ 1.24–1.57, FDi ∼ 0.05−−1.0
α ∼ 1.22–1.34, FDi ∼ 0.05−−0.07
262 E. Bek-Pedersen, R. Gani / Chemical Engineering and Processing 43 (2004) 251–262

When the order of the splits has been determined, the distillation columns, based on the simple and visual driv-
actual positions of the feed and the side draw are determined ing force techniques. Finally, the results appear to confirm
accordingly. The results of the predictions have been verified the theory that separation at the highest driving force is
with rigorous simulation, as is indicated in Tables 4 and 5. the easiest separation and, therefore, should require a near
minimum of energy since energy is needed to create the
Example 3 (Separation of BTX mixture). Another exam- driving force. One limitation of the method is that the two
ple is a classical separation problem, where the mixture of adjacent products from the distillation column must lie on
benzene, toluene and xylene is to be separated. This mix- either side of the maximum driving force.
ture is also to be separated in a column with three product
streams, and Algorithm D2 has been applied to solve this
problem. The column is specified to operate at 10 atm pres- Appendix A. Nomenclature
sure and has 40 stages. Just as in Example 2, the steps of
the algorithm are presented in a schematic form in Table 6. A product composition specification (see Fig. 1)
B product composition specification (see Fig. 1)
C multiplication factor for RRmin to RR
Example 4 (Retrofit design). This is an example of the
D largest driving force
retrofit Algorithm R1. Here we are dealing with a column
Ds relative position of side-draw driving force
of 60 stages, and we have options to feed on stages 33 and
Dx relative position of largest driving force
38. The steps outlined above for Algorithm R1 has been
Dy size of largest driving force
followed, making use of Table 2, and Fig. 2. The results of
F feed (kmol/h)
the example are listed in Table 7. With the given product
Fij , FDi driving force
specifications, and the corresponding driving force, the cor-
HK heavy key compound
responding range of relative volatility was easily identified,
LK light key compound
and three mixtures suitable for separation in this column
RR reflux ratio
were immediately found on this basis. Note that in step 6,
RRmin minimum reflux ratio
different mixtures match the calculated relative volatilities
N number of stages
at different operating pressures.
NF feed stage
NP number of stages
NS side draw stage
4. Conclusions P pressure (atm)
T temperature (K)
An integrated framework for synthesis and design of xi liquid composition of compound i
distillation-based separation systems has been developed yi vapour composition of compound i
and validated. The framework is based on the driving force
approach [3] and further extends it, enabling thereby the Greek letters
solution of a wide range of problems related to distillation α relative volatility
columns. The framework consists of six algorithms that in β relative separability
an integrated manner interactively enable the visual deter- θ resistance factor
mination of the near-optimal (if not optimal) design, as well
as separation sequences together with the corresponding Subscripts
condition of operation for both conventional and complex B, Bot bottom composition
distillation columns. The integrated approach also allows D, Dist distillate composition
the generation of hybrid separation schemes where different Z feed composition
separation techniques are allowed. With this approach, the
only requirements for application of the integrated frame-
work are the co-existing phase composition data. The ap- References
proach requires no rigorous simulation or optimisation. The
methodology not only identifies the feasibility of different [1] E. Bek-Pedersen, M. Hostrup, R. Gani, Eur. Symp. Comput. Aid.
separation techniques for a given separation task, but also in- Process Eng. 10 (2000) 955–960.
dicates the optimum methods of separation. Consequently, it [2] E. Bek-Pedersen, R. Gani, O. Levaux, Comput. Chem. Eng. 24 (2000)
is possible to make early decisions on separation sequences 253–259.
[3] R. Gani, E. Bek-Pedersen, AIChE J. 46 (6) (2000) 1271–1274.
and distillation configurations that are near-optimum solu- [4] PRO-II Users Guide, Simulation Sciences—An Invensys Company,
tions. It is evident that this extended methodology is able to Brea, CA, 2001.
generate near-optimum designs of even very complicated [5] D.W. Tedder, D.F. Rudd, AIChE J. 24 (2) (1978) 303–315.